🔥 推特起爆帖监控
搜索和分析大V账号即将起爆的热帖推文。通过SoPilot插件快速评论和互动,抢先一步占领评论区,你将获得更多的曝光。
终于看到靠谱的日更AI资讯聚合,来自著名的AI Newsletter:latent. space 如果没时间精力,读这个频道就够,不用折腾。 信息源:12个Reddit子版、500多个 Twitter账号 和 24 个 Discord频道,每天累计上万信息筛选精华。 这家专业性靠谱,Ben's bite也很好,但是周更。 地址见评论区,建议收藏
人类 Get 龙虾openclaw新技能! 同事在自己服务器上装 openclaw 接钉钉机器人。找了技术同事帮忙,各种报错,折腾一中午没跑起来。 我说让我的 ai 去干吧。 把服务器 IP 和密码丢给 AI,只说了一句话: “连不上钉钉,你看看啥问题。” 就这一句。没有操作步骤,没有排查思路,没有任何指挥。 然后我半小时后他搞定了。 它自己 SSH 上去,看日志,查配置,翻文档,改文件,重启验证,最后把服务跑起来。 还顺手给我总结了问题清单,这些问题,说难也不难,但排查起来就是烦。 而我只描述了现象,它自己闭环了。 这么算下来在很多场景 AI 是能独立干活的那种了。 它之所以能做到这一点,其实有个很关键的前提,我把官方文档 290 页全部同步进 NotebookLM,每天自动更新。 它排错的时候,会主动去确认最新配置方式。 等于给 AI 配了一本永远最新的操作手册。 很多人觉得 AI 不靠谱,你不给它完整上下文,它当然只能猜,但如果文档齐全,规则明确呢? 然后下面这些, 服务器部署、日志排错、监控告警响应、竞品监控等等等等 凡是有文档可查,有标准流程的事,AI 都能跑完整闭环。 人类只在异常边界出现,烦人的活交给不会烦的队友干吧。
LarryBrain 的赚钱和增长思路 很抓人性 用 OpenClaw 的大众 一片韭菜呀 https://t.co/pCkdCv8hk8
咱就是说,欧洲人办事效率能有夺低呢? 去年年初跟情人打得火热的时候,跟公寓管理员说给我房子再配一副钥匙给他用。 今年年初我已经把他拉黑了删除了伤心了哭过了这辈子不想再见了重新振作开始新生活了,钥匙还没配好呢!🤣😝😤🤷🏻♀️ (为防抬杠:我也是申请新钥匙后才知道,有些房子的钥匙很特别,不能随便找店配;房子是买的不是租的,也得通过property management办公室弄钥匙。)
啊哈??? 美国政府部门推特这么low啊 这种事儿不就是各国庆祝各国的 各国开心各国的😂😂 突然莫名其妙的嘲讽, 显得真的过于low逼了 这是把自媒体账号给了 某官员10后儿子运营了吧
第一次看㍿这个字的时候我都惊了,还以为是谁做的表情包,后来查了一下,发现还有:㌃㌄㌅㌆㌇㌈㌉㌊㌋㌌㌍㌎㌏㌐㌑㌒㌓㌔㌕㌖㌗㌘㌙㌚㌛㌜㌝㌞㌟㌠㌡㌢㌣㌤㌥㌦㌧㌨㌩㌪㌫㌬㌭㌮㌯㌰㌱㌲㌳㌴㌵㌶㌷㌸㌹㌺㌻㌼㌽㌾㌿㍀㍁㍂㍃㍄㍅㍆㍇㍈㍉㍊㍋㍌㍍㍎㍏㍐㍑㍒㍓㍔㍕㍖㍗㍘㍙㍚㍛㍜㍝㍞㍟㍠㍡㍢㍣㍤㍥㍦㍧㍨㍩㍪㍫㍬㍭㍮㍯㍰㍱㍲㍳㍴㍵㍶㍷㍸㍹㍺㍻㍼㍽㍾㍿
Anthropic首席执行官刚刚说过了,AI将在12个月内消灭50%的律师、顾问和金融专业人士: 纯知识类的、形式化的和半形式化的、结构化的语言类工作都会被swiped! https://t.co/zucFtkkE7b
A simple prompt, yet it delivers such outstanding results! A low-quality photo that looks like it was taken by an amateur: a side-profile close-up of a cute Japanese woman (no makeup, 18 years old, slim build) with long black hair and her forehead visible, wearing a thick coat and scarf, walking along a winter beach covered with a thin layer of snow. The atmosphere is cold, and there is no one else around.
最近把我每天都在开的一个小工具开源了,叫 Eve。 我一般拿它记会议和临时想法,后台挂着就行,不用反复点录制。 - 24 小时连续录音 - 自动分段存 WAV - 本地实时转写(可关) https://t.co/HBF1u9wJ5h https://t.co/5uLdDFcHzl
我用 AI Agent 跑 Google Ads,10 分钟从 0 开始建了 2 个 campaign、8 个 ads group,关键词、出价、文案、附加信息全部配好。 本来只是测试一下自己的产品。结果还没正式上线,2 天拿了 3 个转化。 这个产品叫 AdWhiz,一句话解释:给 AI Agent 用的广告投放 API。 它现在只是一个 MCP Server,47 个工具,任何 AI agent 都能通过它操作你的 Google Ads 账户,建计划,调预算,暂停烧钱的词,跑审计,迭代优化。 但真正有意思的是它接入了 OpenClaw 生态: 别的 skill 帮你生成素材、文案、投放策略, AdWhiz 帮你一键上线到 Google Ads, AdWhiz 每天自动审计,给账户打分, Agent 看数据,自动迭代下一版 生成,投放,监测,迭代,全闭环,不用打开 Google Ads 那个狗屎一样的后台,我每次找个配置都要几十分钟。 我自己之前的 13 个 campaign,以前每次 review 要半天。现在每小时自动跑一遍,前天查出 $1300+ 的浪费。 目前 Google Ads 已上线。AI 对话 + Meta Ads 马上发。每天都在发版。 另外搞了个 affiliate,推荐分佣最高 70%,终身。不是开玩笑。
好家伙,这是哪位兄弟,我怎么就没想到呢 来,好东西分享给大家 https://t.co/IZQHnehV2c https://t.co/LbQSoeFwvV
Can you go to bed early tonight? https://t.co/PcbSXz1aQu { "subject": { "description": "An intimate boudoir-style portrait of an adult woman on a bed, leaning forward toward the camera with a calm, soft gaze and a faint smile; dewy skin highlights and a quiet, private bedroom mood, styled like a polished Korean celebrity private shoot while keeping the lingerie boudoir feel.", "mirror_rules": [ "not a mirror selfie", "no mirrored text artifacts" ], "age": "adult (over 21)", "expression": { "eyes": { "look": "gentle, slightly sleepy, direct eye contact", "energy": "soft and intimate", "direction": "looking into the lens" }, "mouth": { "position": "closed lips with a subtle smile", "energy": "warm, relaxed" }, "overall": "quiet confidence, tender and unposed" }, "face": { "preserve_original": true, "makeup": "K-beauty celebrity styling: glass-skin base, soft taupe-brown liner, subtly emphasized aegyo-sal, naturally groomed straight brows, glossy gradient nude-pink lips, minimal blush" }, "hair": { "color": "dark brown to black", "style": "clean center part with a sleek low bun, a few wispy face-framing strands kept on purpose", "effect": "polished yet effortless, lived-in softness" }, "body": { "frame": "slim to average with soft curves", "waist": "gentle taper", "chest": "curvy bust, natural shape", "legs": "mostly out of focus/partially out of frame", "skin": { "visible_areas": [ "face", "neck", "shoulders", "arms", "upper chest", "collarbones" ], "tone": "fair with a cool-neutral undertone", "texture": "smooth and moisturized, fresh skincare slip, not plastic", "lighting_effect": "specular highlights on forehead, cheekbones, nose bridge, lips, collarbones" } }, "pose": { "position": "on a bed, leaning forward toward camera", "base": "hands supporting on the mattress, torso angled forward", "overall": "slight head tilt, shoulders forward, intimate distance" }, "clothing": { "top": { "type": "luxury lace balconette bra", "color": "beige/nude", "details": "delicate lace texture with fine floral pattern, clean premium stitching, straps slipped down to the upper arms (intentional), no logos", "effect": "Korean celebrity private-shoot lingerie: refined, elegant, softly undone" }, "bottom": { "type": "matching lace panties", "color": "beige/nude", "details": "fine lace pattern, minimal seams, no logos" } } }, "accessories": { "jewelry": "minimal fine gold jewelry: tiny hoops or small studs, one thin delicate necklace (logo-free)", "prop": "none" }, "photography": { "camera_style": "high-end digital portrait with a natural, candid boudoir feel", "angle": "slightly above eye-level, close and intimate", "shot_type": "vertical 3/4 portrait, face and upper torso dominant in frame", "aspect_ratio": "9:16", "texture": "clean detail with mild low-light softness and subtle noise", "lighting": "single soft window light from camera-left, cool-toned low-key exposure, gentle shadows", "depth_of_field": "moderate shallow depth, background softly blurred" }, "background": { "setting": "minimal bedroom", "wall_color": "cool gray to off-white", "elements": [ "simple closet/door panels", "rumpled bed sheets", "soft out-of-focus bedding" ], "atmosphere": "quiet, private, early-morning stillness", "lighting": "dim ambient room light with cool window illumination" }, "the_vibe": { "energy": "soft and hushed", "mood": "intimate, calm, slightly melancholic", "aesthetic": "realistic social-photo boudoir, minimal and moody", "authenticity": "unposed, lived-in imperfections, slight framing drift", "intimacy": "close personal distance, gentle eye contact", "story": "As if a stylist quietly prepped her for a private shoot in a silent room, but the moment stayed human: a few loose strands, a breath held between the cool window light and the warm skin sheen.", "caption_energy": "quiet tease, tender confidence" }, "constraints": { "must_keep": [ "cool gray-blue color grade", "soft window light from the left", "dewy glossy skin highlights", "close intimate framing", "leaning-forward pose on rumpled bed", "minimal bedroom background with door/closet lines", "beige/nude lace lingerie with straps slipped down", "realistic candid feel (not studio glamour)" ], "avoid": [ "explicit nudity", "visible nipples", "graphic genital detail", "heavy glam makeup", "harsh flash lighting", "overly perfect symmetrical composition", "cartoon/anime styling", "text, logos, watermarks" ] }, "negative_prompt": [ "explicit nude", "nipples visible", "graphic detail", "pornographic", "underage", "heavy airbrushing", "plastic skin", "harsh flash", "overexposed highlights", "high saturation", "anime", "CGI", "watermark", "logo", "text" ] }
最近被 gpt 5.3 codex 不说人话搞的有点暴躁, 阅读体验贼差, 特别是读 plan 的 ..... 事无巨细 + 黑话频出 我试过通过在 agents md 里写各种提示词来改善沟通, 但 codex 5.3 就是死活不听话 接着我又想到一个很严肃的问题 「我真的有必要去“阅读”事无巨细 + 黑话频出的plan 嘛? 输入 codex 和输出 codex 这个流程是不是都可以简单的优化一下? 我是不是可以找个小助手帮我解决? 我只看重点?」 所以, 我火速试了一下在 openclaw 里面去调用 codex 来规划, 这里我用的是 @MiniMax_AI 的 M2.5, 你可以选择你熟悉的模型, 把一些输入偏好和阅读偏好放入龙虾中, 比如我更希望用表格呈现. 用 openclaw tui 快速跑通流程, 如视频, 左侧是优化版本的 openclaw给我的报告, 右边是 codex 默认的报告. 下一步就可以接入飞书或者discord了.
这几天一直在做设计 Agent 终于在 Nanabonana 2 上线的时候赶出来了! 内嵌在我的开源 Claude Code 桌面端 Codepilot里。 能选参数、能连续编辑、还有素材库帮你管提示词和收藏图片,我的提示和思路后续都会内置到里面。 ------ 录个视频介绍一下: 支持Nano Banana、 Pro 和 2三个模型生图 参数自由选择:想要什么风格、什么比例,自己定。 连续编辑:反复调整细节的时候特别有用。 ## 素材库 内置素材库可以收藏用过的好图,也能快速找到之前写过的提示词。下次想生类似风格的图,直接从素材库调,效率翻倍。 ## 设置方法 5 步搞定: 1. 去 Google AI Studio 申请一个 API Key 2. 打开 Codepilot 设置,找到服务商 3. 点进 Google Gemini (Image) 4. 把 Key 填进去 5. 回到聊天页面,切换到 image-Agent 模式
我不认可藏师傅这句话 1. 创业重要的是活下去,是业务,是现金流,不是会不会openclaw 2. 商业落地都不一定是最fomo的,甚至可以说99%的创业都不是因为这个团队技术世界第一。 3. 能用AI来优化自己的业务形态固然很好,但是系统的稳定永远是第一要义 https://t.co/6IwoCBjOxD
Using nano banana 2 in @ImagineArt_X { "subject": "Young female with long brown hair wearing a form-fitting, long-sleeved, short-legged white onesie patterned with peaches.", "pose": "Standing in profile, facing left, torso twisted to look back at the mirror. Right hand holds a black smartphone to take a selfie. Back is arched, emphasizing posterior volume. Lips puckered.", "environment": "Bathroom setting. A white paneled door with silver hardware is directly behind. A white sink basin and chrome faucet occupy the bottom foreground. A white light switch is visible on the left wall.", "camera": "Mirror selfie perspective. Shot from chest height, cropping the subject from mid-thigh to just above the head.", "lighting": "Standard indoor artificial overhead lighting, casting soft shadows on the door panels and bright reflections on the metallic faucet.", "mood_and_expression": "Casual, posed selfie posture with puckered lips.", "style_and_realism": "Unedited, realistic smartphone photograph.", "colors_and_tone": "Warm overall tone dominated by whites, warm skin hues, peach/orange accents on the garment, and metallic silver/chrome.", "quality_and_technical_details": "Typical modern smartphone camera resolution with slight indoor grain and realistic fabric tension.", "aspect_ratio_and_output": "3:4", "controlnet": { "pose_control": { "model_type": "openpose_full", "purpose": "Preserve exact skeletal alignment, torso orientation, and anatomical proportions without altering mass distribution", "constraints": "No limb length change, no torso compression, no chest volume reduction, no posture modification", "recommended_weight": 1.1 }, "depth_control": { "model_type": "depth_midas", "purpose": "Preserve accurate 3D spatial depth, volumetric projection, and foreground-background separation", "constraints": "No depth flattening, no torso narrowing, no volumetric shrinkage, no perspective alteration", "recommended_weight": 1.15 } }, "negative_prompt": "stylization, body reshaping, anatomical distortion, breast reduction, volume compression, proportion changes, depth flattening, beautification filters, symmetry correction, torso slimming, perspective alteration" }
昨天NVDA说的比较含蓄了,懂的都懂,但是喊空容易挨骂,有做日内的朋友已经掏出了两单的日内put了,100%收益很简单。 杀完call的买方之后,可能会回到熟悉的慢涨节奏里,就不要继续玩了,换下一个标的。 为什么有这个经验呢,因为当年财报远超预期的时候我就被NVDA的财报末日call杀过IV,涨了很多,IV被提前买起来了,然后亏钱。
传统软件开发的时候,有复杂的测试和质量保证流程,因为过去基于光盘的软件发布逻辑,修复 bug 是非常非常难的。 过渡到 CI/CD 时代,各大公司开始大幅裁撤测试团队(著名的有微软和 google),因为软件可以方便快速的在线修复 bug,高昂的测试团队成本变得不合算 现在来到 Vibe 时代,应该会有一次新的变化,可能是:“review 代码不重要了” “程序员要放弃对代码的掌控力” “测试变得愈加重要” “human in the loop 的相关哲学更加重要” “AI 自动化测试”
Nano banana pro or Grok? https://t.co/OrTsnek15G { "subject": { "description": "A cute Japanese-style maid cosplay selfie of a young woman, close and intimate, taken with a handheld front-facing smartphone camera from a slightly high angle.", "mirror_rules": "Front-camera selfie (no mirror). Keep natural selfie perspective and slight wide-angle distortion. No mirrored/reflected text. Phone stays out of frame but the extended arm/shoulder angle should feel real.", "age": "early 20s", "expression": { "eyes": { "look": "soft, slightly downcast gaze toward the camera", "energy": "calm, gentle, a little shy", "direction": "slightly downward and to the side" }, "mouth": { "position": "lips softly closed", "energy": "quiet, restrained sweetness" }, "overall": "serene, slightly pensive kawaii mood" }, "face": { "preserve_original": true, "makeup": "light Japanese makeup: smooth base, subtle eyeliner, long lashes, natural straight brows, soft blush, natural rosy-pink lips" }, "hair": { "color": "dark brown", "style": "low loose bun gathered to the back/side with face-framing strands and wispy bangs", "effect": "slightly messy, lived-in softness with a few stray hairs" }, "body": { "frame": "slim petite", "waist": "slim", "chest": "modest", "legs": "not visible", "skin": { "visible_areas": "face, neck, upper chest/cleavage area, upper arms", "tone": "fair", "texture": "silky and smooth with gentle softness (beauty-filtered but not plastic)", "lighting_effect": "warm soft highlight on cheeks and forehead, minimal harsh shadows" } }, "pose": { "position": "close to the camera, shoulders slightly angled", "base": "one arm extended forward holding the phone (out of frame)", "overall": "slightly tilted selfie posture, intimate and casual" }, "clothing": { "top": { "type": "classic black-and-white maid dress bodice with white ruffled collar and apron bib", "color": "black base with white ruffles", "details": "white frilled neckline, white ruffled trim, black ribbon bow at the collar, puff sleeves with white frills", "effect": "clean, crisp contrast; cute and traditional maid aesthetic" }, "bottom": { "type": "maid dress skirt with white apron", "color": "black skirt with white apron", "details": "white apron with ruffled edges", "effect": "soft fabric folds, slightly rounded silhouette" } } }, "accessories": { "headwear": "none", "jewelry": "none", "device": "smartphone front camera (out of frame), handheld selfie feel", "prop": "none" }, "photography": { "camera_style": "realistic social media smartphone selfie, front-facing camera aesthetic", "angle": "slightly high angle, close distance, subtle wide-angle perspective", "shot_type": "tight close-up to upper-body (head and torso dominate the frame)", "aspect_ratio": "9:16 vertical", "texture": "soft beauty smoothing, mild haze, fine grain, gentle sharpening on eyes, slightly imperfect focus", "lighting": "indoor warm soft light, diffused and even, no harsh rim light", "depth_of_field": "shallow to moderate, face sharpest with softly softened background" }, "background": { "setting": "simple indoor room", "wall_color": "light beige / off-white", "elements": [ "plain wall", "minimal distractions", "subtle shadow gradient" ], "atmosphere": "quiet, cozy, private selfie moment", "lighting": "soft warm ambient light with gentle falloff" }, "the_vibe": { "energy": "soft and intimate", "mood": "kawaii, calm, slightly wistful", "aesthetic": "Japanese cute maid cosplay with realistic social media authenticity", "authenticity": "handheld front-camera feel, casual tilt, tiny imperfections like stray hairs and mild smoothing", "intimacy": "close personal distance, like a secret selfie sent to someone trusted", "story": "She had just tied her hair into a loose bun, put on the maid outfit, and took a quick front-camera shot before stepping out of the room.", "caption_energy": "quiet cute, understated sweetness" }, "constraints": { "must_keep": [ "Japanese cute maid cosplay", "black-and-white ruffled maid dress with black ribbon bow", "dark brown hair in a low loose bun with face-framing strands", "soft calm downcast gaze", "handheld front-facing smartphone selfie perspective", "plain light wall background", "warm soft indoor lighting", "soft beauty smoothing with slight natural imperfections" ], "avoid": [ "mirror selfie reflection", "heavy glamour makeup", "dramatic colored lights", "busy background objects", "strong film scratches or vintage film borders", "visible logos or readable text" ] }, "negative_prompt": [ "lowres", "overexposed", "harsh shadows", "plastic skin", "over-sharpened", "distorted face", "extra fingers", "deformed hands", "bad anatomy", "cross-eye", "open mouth teeth", "greasy hair", "busy background", "logos", "readable text", "watermark" ] }
Anthropic 向开源贡献者免费送半年 Claude Max Anthropic 推出"Claude for Open Source"计划,向开源社区的维护者和核心贡献者提供 6 个月免费的 Claude Max 20x 订阅。 Claude Max 20x 是目前 Claude 最高级别的付费方案,正常订阅价格不菲,这次相当于直接把最强工具送到开源开发者手里。 申请门槛不算低:你需要是 GitHub 星标 5000 以上或 NPM 月下载量超百万的公开项目的主要维护者或核心团队成员,且近三个月内有活跃的代码提交、版本发布或 PR 审核记录。 不过 Anthropic 也留了口子:如果你维护的项目虽然不那么显眼,但整个生态都在默默依赖它,同样可以申请。 项目采用滚动审核,总共接受最多一万名贡献者。 这一举措的意图很明确:让写代码的人用上最好的 AI 编程工具,同时把 Claude 更深地植入开发者的日常工作流。 申请地址:https://t.co/Kmamr5oVHJ
即梦 Seedance 2.0 的提示词,我踩了一整个春节的坑。 ❌写"电影感"没用,写"UnrealEngine5渲染+工业光魔级VFX"才有用。 ❌写"光线好"没用,光源→光行为→色调三层分开写才有用。 ❌写"氛围统一"没用,"每帧都有薄雾弥散效果"这种大气连贯声明才有用。 全部整理成了一个开源 Skill,结构化的、可以直接喂给 Claude/GPT 当系统提示词用: ✅ 10 种能力的提示词模板 ✅ 4 种分镜结构(基础/时间戳/快切/史诗) ✅ 运镜三级进阶 + 200 词库 ✅ 9 格分镜图一键批量生成 ✅ 超 15 秒视频分段拼接方案 你不需要懂提示词工程,复制粘贴就能出片。 源码仓库跟参考资料见评论 👇
Claude Code 新增"自动记忆"功能 https://t.co/24zag047Ru Anthropic 为其命令行编码工具 Claude Code 上线了自动记忆(Auto Memory)功能。 开启后,Claude 会在工作过程中自动记录项目中学到的上下文:比如构建命令、调试技巧、代码风格偏好、架构要点等,并在下次会话中自动调用,不需要用户手动记录任何东西。 此前 Claude Code 已经支持 https://t.co/k5gfrsJiDu 文件,但那是用户写给 Claude 的指令。 新增的 https://t.co/wdYpQc1q03 则相反,是 Claude 自己维护的"笔记本"。 用户也可以直接告诉 Claude"记住我们用 pnpm 而不是 npm",它就会写入记忆文件。 技术细节上,每个项目的记忆存储在本地目录 ~/.claude/projects/ 下,会话启动时自动加载 https://t.co/YAPv7h70sj 的前 200 行,更详细的内容按需读取。 该功能默认开启,可通过 /memory 命令或配置文件关闭。 这个功能解决的是开发者反复向 AI 解释项目背景的痛点:用得越久,Claude 对项目越熟悉。 相关文档:https://t.co/7nNFKqDkHW
这两天深度使用了 MaxClaw,是我一次使用 Claw 系列产品。 主要做了一件事,把 AI 论文 GUI 阅读器做成 Agentic 论文阅读伙伴。 首先,根据我的原本的论文 AI 阅读 app,让 MaxClaw 做出一套 CLI 工具。 设计 CLI 的动作时,我要求使用 ACI 的设计原则,量少简洁,包含基本的 find, digest, analyze, report 即可。 在简单动作的基础上,MaxClaw 可以自由把基本动作以不同方式组合起来形成复杂 workflow,以完成难度更高的任务。 之前对 OpenClaw 有太多安全方面的担忧,一直没有使用,但跑在云端的 MaxClaw 的形态让距离产生美,在同样的 personal agent 的体验下,完美消减安全担忧。 Kudos to @MiniMax_AI @MiniMaxAgent , 真的非常棒!
Block大裁员:Jack Dorsey 砍掉近半员工,称 AI 改变了公司运营方式 Block(前身为 Square)CEO Jack Dorsey 宣布裁员约 4000 人,将公司从超过 1 万人缩减至不到 6000 人,裁员比例接近一半。这是 Block 历史上规模最大的一次裁员。 Jack 在致全员信中强调,裁员并非因为经营困难——公司毛利润仍在增长,客户规模持续扩大,盈利能力也在改善。他给出的理由是:AI工具配合更精简扁平的团队,正在从根本上改变公司的运作方式,而这种变化还在加速。 他表示自己在“逐步缓慢裁员”和“一次到位”之间选择了后者,认为反复裁员对士气和信任的伤害更大。被裁员工将获得至少 20 周薪资(加每年工龄1周)、股权归属延续到 5 月底、6 个月医保及 5000 美元过渡补贴。 Jack 在信中明确了 Block 的未来方向: 以AI为核心重建公司运营,并让客户能够直接利用Block的能力自行构建功能。 Block 旗下拥有 Cash App 和 Square 两大支付产品,此次裁员是科技行业“AI驱动组织瘦身”趋势中规模最大的案例之一,后续值得关注这种激进转型能否兑现 Jack 的承诺。
昨晚谷歌正式发布了 Nano Banana 2!🍌 基于 Gemini Flash 模型做推理,相比 Nano Banana Pro 的使用成本下降 50% 以上。 在成本下降的同时,模型的文字能力大幅增强,且生成速度更快 同事用它刻了一座石碑...效果极其震撼 模型使用方式: 1. Gemini 官方可以体验,提供2k分辨率,压缩画质,但好在每天有一些额度可以免费体验 2. Labnana 与官方深度合作,提供4k分辨率,无损高清画质,批量生成,API 接入,Skills 调用等服务。每天签到可免费体验,也可以充值积分解锁无限使用。 Labnana 是全球首个 Nano Banana Pro 套壳独立站,全部按照官方规范,去掉了一切的负优化,专注把🍌这个 Agentic 生图模型的实力发挥到最大,我们在昨晚凌晨也在第一时间支持了 Nano Banana 2,欢迎体验。
2026 年 “Harness Engineering” 这个词要火。 “Harness” 这个词,字面意思是“马具”,就是套在马身上、让人能控制马匹方向和力量的那套装备。 用在 AI 编程的语境里,它的比喻再贴切不过:AI Agent 就像一匹动力十足但不太守规矩的马,而 Harness 就是那套让它既能跑得快、又不会跑偏的缰绳和马鞍。 过去三年,三个阶段: 1. Prompt Engineering(2023-2024):关注“怎么跟 AI 说话” 精心设计一段提示词,希望模型给出理想输出。Prompt Engineering 是优化一次性的输入-输出对。 局限很明显:一条消息能塞的信息有限,任务一复杂就失控。 2. Context Engineering(2025):关注“给 AI 看什么信息” 不再只盯措辞,而是设计整个信息环境:系统提示、对话历史、记忆、RAG 检索结果、工具调用输出。 3. Harness Engineering(2026):关注“构建什么环境让 AI 工作,这个环境如何保证它的产出是可靠的” 比 Context Engineering 更进一步,不仅管理输入给模型的信息,还包括模型之外的整个执行环境。 现在问题是,“Harness Engineering”中文怎么说?
Nano Banana 2 图像生成模型正式发布,速度更快、价格更低 谷歌又开始大撒币了,Flow 里面 Nano Banana 2 的图片生成完全免费! - 依旧支持世界知识和搜索 - 更加精确的文本渲染和翻译 - 支持最多 14 个主体的一致性保持 - 更强的指令遵循能力 - 更高的光影、纹理和细节表现 - 非常快的生成速度 - API 价格仅为 Pro 的一半,1K 图片 $0.067 - 在谷歌所有平台都已经上线
为什么管理 AI Agent 不能像管理代码一样简单? GitAgent提了一个新的开放标准。 一份定义,到处运行,只需要写三个核心文件(agent.yaml、https://t.co/PdGIg5g6Sx、https://t.co/KtWZtNnyC3),然后一行命令导出到任何平台: $ gitagent export -f claude-code # 导出给 Claude $ gitagent export -f openai # 导出给 OpenAI $ gitagent export -f crewai # 导出给 CrewAI 让 Agent 变得可追溯、可协作、可复用。 还没用,但看着理念不错,网址见评论。